| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * PSCI CPU idle driver. |
| 4 | * |
| 5 | * Copyright (C) 2019 ARM Ltd. |
| 6 | * Author: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> |
| 7 | */ |
| 8 | |
| 9 | #define pr_fmt(fmt) "CPUidle PSCI: " fmt |
| 10 | |
| 11 | #include <linux/cpuhotplug.h> |
| 12 | #include <linux/cpu_cooling.h> |
| 13 | #include <linux/cpuidle.h> |
| 14 | #include <linux/cpumask.h> |
| 15 | #include <linux/cpu_pm.h> |
| 16 | #include <linux/kernel.h> |
| 17 | #include <linux/module.h> |
| 18 | #include <linux/of.h> |
| 19 | #include <linux/device/faux.h> |
| 20 | #include <linux/psci.h> |
| 21 | #include <linux/pm_domain.h> |
| 22 | #include <linux/pm_runtime.h> |
| 23 | #include <linux/slab.h> |
| 24 | #include <linux/string.h> |
| 25 | #include <linux/syscore_ops.h> |
| 26 | |
| 27 | #include <asm/cpuidle.h> |
| 28 | #include <trace/events/power.h> |
| 29 | |
| 30 | #include "cpuidle-psci.h" |
| 31 | #include "dt_idle_states.h" |
| 32 | #include "dt_idle_genpd.h" |
| 33 | |
| 34 | struct psci_cpuidle_data { |
| 35 | u32 *psci_states; |
| 36 | struct device *dev; |
| 37 | }; |
| 38 | |
| 39 | struct psci_cpuidle_domain_state { |
| 40 | struct generic_pm_domain *pd; |
| 41 | unsigned int state_idx; |
| 42 | u32 state; |
| 43 | }; |
| 44 | |
| 45 | static DEFINE_PER_CPU_READ_MOSTLY(struct psci_cpuidle_data, psci_cpuidle_data); |
| 46 | static DEFINE_PER_CPU(struct psci_cpuidle_domain_state, psci_domain_state); |
| 47 | static bool psci_cpuidle_use_syscore; |
| 48 | |
| 49 | void psci_set_domain_state(struct generic_pm_domain *pd, unsigned int state_idx, |
| 50 | u32 state) |
| 51 | { |
| 52 | struct psci_cpuidle_domain_state *ds = this_cpu_ptr(&psci_domain_state); |
| 53 | |
| 54 | ds->pd = pd; |
| 55 | ds->state_idx = state_idx; |
| 56 | ds->state = state; |
| 57 | } |
| 58 | |
| 59 | static inline void psci_clear_domain_state(void) |
| 60 | { |
| 61 | __this_cpu_write(psci_domain_state.state, 0); |
| 62 | } |
| 63 | |
| 64 | static __cpuidle int __psci_enter_domain_idle_state(struct cpuidle_device *dev, |
| 65 | struct cpuidle_driver *drv, int idx, |
| 66 | bool s2idle) |
| 67 | { |
| 68 | struct psci_cpuidle_data *data = this_cpu_ptr(&psci_cpuidle_data); |
| 69 | u32 *states = data->psci_states; |
| 70 | struct device *pd_dev = data->dev; |
| 71 | struct psci_cpuidle_domain_state *ds; |
| 72 | u32 state = states[idx]; |
| 73 | int ret; |
| 74 | |
| 75 | ret = cpu_pm_enter(); |
| 76 | if (ret) |
| 77 | return -1; |
| 78 | |
| 79 | /* Do runtime PM to manage a hierarchical CPU toplogy. */ |
| 80 | if (s2idle) |
| 81 | dev_pm_genpd_suspend(dev: pd_dev); |
| 82 | else |
| 83 | pm_runtime_put_sync_suspend(dev: pd_dev); |
| 84 | |
| 85 | ds = this_cpu_ptr(&psci_domain_state); |
| 86 | if (ds->state) |
| 87 | state = ds->state; |
| 88 | |
| 89 | trace_psci_domain_idle_enter(dev->cpu, state, s2idle); |
| 90 | ret = psci_cpu_suspend_enter(state) ? -1 : idx; |
| 91 | trace_psci_domain_idle_exit(dev->cpu, state, s2idle); |
| 92 | |
| 93 | if (s2idle) |
| 94 | dev_pm_genpd_resume(dev: pd_dev); |
| 95 | else |
| 96 | pm_runtime_get_sync(dev: pd_dev); |
| 97 | |
| 98 | cpu_pm_exit(); |
| 99 | |
| 100 | /* Correct domain-idlestate statistics if we failed to enter. */ |
| 101 | if (ret == -1 && ds->state) |
| 102 | pm_genpd_inc_rejected(genpd: ds->pd, state_idx: ds->state_idx); |
| 103 | |
| 104 | /* Clear the domain state to start fresh when back from idle. */ |
| 105 | psci_clear_domain_state(); |
| 106 | return ret; |
| 107 | } |
| 108 | |
| 109 | static int psci_enter_domain_idle_state(struct cpuidle_device *dev, |
| 110 | struct cpuidle_driver *drv, int idx) |
| 111 | { |
| 112 | return __psci_enter_domain_idle_state(dev, drv, idx, s2idle: false); |
| 113 | } |
| 114 | |
| 115 | static int psci_enter_s2idle_domain_idle_state(struct cpuidle_device *dev, |
| 116 | struct cpuidle_driver *drv, |
| 117 | int idx) |
| 118 | { |
| 119 | return __psci_enter_domain_idle_state(dev, drv, idx, s2idle: true); |
| 120 | } |
| 121 | |
| 122 | static int psci_idle_cpuhp_up(unsigned int cpu) |
| 123 | { |
| 124 | struct device *pd_dev = __this_cpu_read(psci_cpuidle_data.dev); |
| 125 | |
| 126 | if (pd_dev) { |
| 127 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
| 128 | pm_runtime_get_sync(dev: pd_dev); |
| 129 | else |
| 130 | dev_pm_genpd_resume(dev: pd_dev); |
| 131 | } |
| 132 | |
| 133 | return 0; |
| 134 | } |
| 135 | |
| 136 | static int psci_idle_cpuhp_down(unsigned int cpu) |
| 137 | { |
| 138 | struct device *pd_dev = __this_cpu_read(psci_cpuidle_data.dev); |
| 139 | |
| 140 | if (pd_dev) { |
| 141 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
| 142 | pm_runtime_put_sync(dev: pd_dev); |
| 143 | else |
| 144 | dev_pm_genpd_suspend(dev: pd_dev); |
| 145 | |
| 146 | /* Clear domain state to start fresh at next online. */ |
| 147 | psci_clear_domain_state(); |
| 148 | } |
| 149 | |
| 150 | return 0; |
| 151 | } |
| 152 | |
| 153 | static void psci_idle_syscore_switch(bool suspend) |
| 154 | { |
| 155 | bool cleared = false; |
| 156 | struct device *dev; |
| 157 | int cpu; |
| 158 | |
| 159 | for_each_possible_cpu(cpu) { |
| 160 | dev = per_cpu_ptr(&psci_cpuidle_data, cpu)->dev; |
| 161 | |
| 162 | if (dev && suspend) { |
| 163 | dev_pm_genpd_suspend(dev); |
| 164 | } else if (dev) { |
| 165 | dev_pm_genpd_resume(dev); |
| 166 | |
| 167 | /* Account for userspace having offlined a CPU. */ |
| 168 | if (pm_runtime_status_suspended(dev)) |
| 169 | pm_runtime_set_active(dev); |
| 170 | |
| 171 | /* Clear domain state to re-start fresh. */ |
| 172 | if (!cleared) { |
| 173 | psci_clear_domain_state(); |
| 174 | cleared = true; |
| 175 | } |
| 176 | } |
| 177 | } |
| 178 | } |
| 179 | |
| 180 | static int psci_idle_syscore_suspend(void *data) |
| 181 | { |
| 182 | psci_idle_syscore_switch(suspend: true); |
| 183 | return 0; |
| 184 | } |
| 185 | |
| 186 | static void psci_idle_syscore_resume(void *data) |
| 187 | { |
| 188 | psci_idle_syscore_switch(suspend: false); |
| 189 | } |
| 190 | |
| 191 | static const struct syscore_ops psci_idle_syscore_ops = { |
| 192 | .suspend = psci_idle_syscore_suspend, |
| 193 | .resume = psci_idle_syscore_resume, |
| 194 | }; |
| 195 | |
| 196 | static struct syscore psci_idle_syscore = { |
| 197 | .ops = &psci_idle_syscore_ops, |
| 198 | }; |
| 199 | |
| 200 | static void psci_idle_init_syscore(void) |
| 201 | { |
| 202 | if (psci_cpuidle_use_syscore) |
| 203 | register_syscore(syscore: &psci_idle_syscore); |
| 204 | } |
| 205 | |
| 206 | static void psci_idle_init_cpuhp(void) |
| 207 | { |
| 208 | int err; |
| 209 | |
| 210 | err = cpuhp_setup_state_nocalls(state: CPUHP_AP_CPU_PM_STARTING, |
| 211 | name: "cpuidle/psci:online" , |
| 212 | startup: psci_idle_cpuhp_up, |
| 213 | teardown: psci_idle_cpuhp_down); |
| 214 | if (err) |
| 215 | pr_warn("Failed %d while setup cpuhp state\n" , err); |
| 216 | } |
| 217 | |
| 218 | static __cpuidle int psci_enter_idle_state(struct cpuidle_device *dev, |
| 219 | struct cpuidle_driver *drv, int idx) |
| 220 | { |
| 221 | u32 *state = __this_cpu_read(psci_cpuidle_data.psci_states); |
| 222 | |
| 223 | return CPU_PM_CPU_IDLE_ENTER_PARAM_RCU(psci_cpu_suspend_enter, idx, state[idx]); |
| 224 | } |
| 225 | |
| 226 | static const struct of_device_id psci_idle_state_match[] = { |
| 227 | { .compatible = "arm,idle-state" , |
| 228 | .data = psci_enter_idle_state }, |
| 229 | { }, |
| 230 | }; |
| 231 | |
| 232 | int psci_dt_parse_state_node(struct device_node *np, u32 *state) |
| 233 | { |
| 234 | int err = of_property_read_u32(np, propname: "arm,psci-suspend-param" , out_value: state); |
| 235 | |
| 236 | if (err) { |
| 237 | pr_warn("%pOF missing arm,psci-suspend-param property\n" , np); |
| 238 | return err; |
| 239 | } |
| 240 | |
| 241 | if (!psci_power_state_is_valid(state: *state)) { |
| 242 | pr_warn("Invalid PSCI power state %#x\n" , *state); |
| 243 | return -EINVAL; |
| 244 | } |
| 245 | |
| 246 | return 0; |
| 247 | } |
| 248 | |
| 249 | static int psci_dt_cpu_init_topology(struct cpuidle_driver *drv, |
| 250 | struct psci_cpuidle_data *data, |
| 251 | unsigned int state_count, int cpu) |
| 252 | { |
| 253 | /* Currently limit the hierarchical topology to be used in OSI mode. */ |
| 254 | if (!psci_has_osi_support()) |
| 255 | return 0; |
| 256 | |
| 257 | data->dev = dt_idle_attach_cpu(cpu, name: "psci" ); |
| 258 | if (IS_ERR_OR_NULL(ptr: data->dev)) |
| 259 | return PTR_ERR_OR_ZERO(ptr: data->dev); |
| 260 | |
| 261 | psci_cpuidle_use_syscore = true; |
| 262 | |
| 263 | /* |
| 264 | * Using the deepest state for the CPU to trigger a potential selection |
| 265 | * of a shared state for the domain, assumes the domain states are all |
| 266 | * deeper states. On PREEMPT_RT the hierarchical topology is limited to |
| 267 | * s2ram and s2idle. |
| 268 | */ |
| 269 | drv->states[state_count - 1].enter_s2idle = psci_enter_s2idle_domain_idle_state; |
| 270 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
| 271 | drv->states[state_count - 1].enter = psci_enter_domain_idle_state; |
| 272 | |
| 273 | return 0; |
| 274 | } |
| 275 | |
| 276 | static int psci_dt_cpu_init_idle(struct device *dev, struct cpuidle_driver *drv, |
| 277 | struct device_node *cpu_node, |
| 278 | unsigned int state_count, int cpu) |
| 279 | { |
| 280 | int i, ret = 0; |
| 281 | u32 *psci_states; |
| 282 | struct device_node *state_node; |
| 283 | struct psci_cpuidle_data *data = per_cpu_ptr(&psci_cpuidle_data, cpu); |
| 284 | |
| 285 | state_count++; /* Add WFI state too */ |
| 286 | psci_states = devm_kcalloc(dev, n: state_count, size: sizeof(*psci_states), |
| 287 | GFP_KERNEL); |
| 288 | if (!psci_states) |
| 289 | return -ENOMEM; |
| 290 | |
| 291 | for (i = 1; i < state_count; i++) { |
| 292 | state_node = of_get_cpu_state_node(cpu_node, index: i - 1); |
| 293 | if (!state_node) |
| 294 | break; |
| 295 | |
| 296 | ret = psci_dt_parse_state_node(np: state_node, state: &psci_states[i]); |
| 297 | of_node_put(node: state_node); |
| 298 | |
| 299 | if (ret) |
| 300 | return ret; |
| 301 | |
| 302 | pr_debug("psci-power-state %#x index %d\n" , psci_states[i], i); |
| 303 | } |
| 304 | |
| 305 | if (i != state_count) |
| 306 | return -ENODEV; |
| 307 | |
| 308 | /* Initialize optional data, used for the hierarchical topology. */ |
| 309 | ret = psci_dt_cpu_init_topology(drv, data, state_count, cpu); |
| 310 | if (ret < 0) |
| 311 | return ret; |
| 312 | |
| 313 | /* Idle states parsed correctly, store them in the per-cpu struct. */ |
| 314 | data->psci_states = psci_states; |
| 315 | return 0; |
| 316 | } |
| 317 | |
| 318 | static int psci_cpu_init_idle(struct device *dev, struct cpuidle_driver *drv, |
| 319 | unsigned int cpu, unsigned int state_count) |
| 320 | { |
| 321 | struct device_node *cpu_node; |
| 322 | int ret; |
| 323 | |
| 324 | /* |
| 325 | * If the PSCI cpu_suspend function hook has not been initialized |
| 326 | * idle states must not be enabled, so bail out |
| 327 | */ |
| 328 | if (!psci_ops.cpu_suspend) |
| 329 | return -EOPNOTSUPP; |
| 330 | |
| 331 | cpu_node = of_cpu_device_node_get(cpu); |
| 332 | if (!cpu_node) |
| 333 | return -ENODEV; |
| 334 | |
| 335 | ret = psci_dt_cpu_init_idle(dev, drv, cpu_node, state_count, cpu); |
| 336 | |
| 337 | of_node_put(node: cpu_node); |
| 338 | |
| 339 | return ret; |
| 340 | } |
| 341 | |
| 342 | static void psci_cpu_deinit_idle(int cpu) |
| 343 | { |
| 344 | struct psci_cpuidle_data *data = per_cpu_ptr(&psci_cpuidle_data, cpu); |
| 345 | |
| 346 | dt_idle_detach_cpu(dev: data->dev); |
| 347 | psci_cpuidle_use_syscore = false; |
| 348 | } |
| 349 | |
| 350 | static int psci_idle_init_cpu(struct device *dev, int cpu) |
| 351 | { |
| 352 | struct cpuidle_driver *drv; |
| 353 | struct device_node *cpu_node; |
| 354 | const char *enable_method; |
| 355 | int ret = 0; |
| 356 | |
| 357 | cpu_node = of_cpu_device_node_get(cpu); |
| 358 | if (!cpu_node) |
| 359 | return -ENODEV; |
| 360 | |
| 361 | /* |
| 362 | * Check whether the enable-method for the cpu is PSCI, fail |
| 363 | * if it is not. |
| 364 | */ |
| 365 | enable_method = of_get_property(node: cpu_node, name: "enable-method" , NULL); |
| 366 | if (!enable_method || (strcmp(enable_method, "psci" ))) |
| 367 | ret = -ENODEV; |
| 368 | |
| 369 | of_node_put(node: cpu_node); |
| 370 | if (ret) |
| 371 | return ret; |
| 372 | |
| 373 | drv = devm_kzalloc(dev, size: sizeof(*drv), GFP_KERNEL); |
| 374 | if (!drv) |
| 375 | return -ENOMEM; |
| 376 | |
| 377 | drv->name = "psci_idle" ; |
| 378 | drv->owner = THIS_MODULE; |
| 379 | drv->cpumask = (struct cpumask *)cpumask_of(cpu); |
| 380 | |
| 381 | /* |
| 382 | * PSCI idle states relies on architectural WFI to be represented as |
| 383 | * state index 0. |
| 384 | */ |
| 385 | drv->states[0].enter = psci_enter_idle_state; |
| 386 | drv->states[0].exit_latency = 1; |
| 387 | drv->states[0].target_residency = 1; |
| 388 | drv->states[0].power_usage = UINT_MAX; |
| 389 | strscpy(drv->states[0].name, "WFI" ); |
| 390 | strscpy(drv->states[0].desc, "ARM WFI" ); |
| 391 | |
| 392 | /* |
| 393 | * If no DT idle states are detected (ret == 0) let the driver |
| 394 | * initialization fail accordingly since there is no reason to |
| 395 | * initialize the idle driver if only wfi is supported, the |
| 396 | * default archictectural back-end already executes wfi |
| 397 | * on idle entry. |
| 398 | */ |
| 399 | ret = dt_init_idle_driver(drv, matches: psci_idle_state_match, start_idx: 1); |
| 400 | if (ret <= 0) |
| 401 | return ret ? : -ENODEV; |
| 402 | |
| 403 | /* |
| 404 | * Initialize PSCI idle states. |
| 405 | */ |
| 406 | ret = psci_cpu_init_idle(dev, drv, cpu, state_count: ret); |
| 407 | if (ret) { |
| 408 | pr_err("CPU %d failed to PSCI idle\n" , cpu); |
| 409 | return ret; |
| 410 | } |
| 411 | |
| 412 | ret = cpuidle_register(drv, NULL); |
| 413 | if (ret) |
| 414 | goto deinit; |
| 415 | |
| 416 | cpuidle_cooling_register(drv); |
| 417 | |
| 418 | return 0; |
| 419 | deinit: |
| 420 | psci_cpu_deinit_idle(cpu); |
| 421 | return ret; |
| 422 | } |
| 423 | |
| 424 | /* |
| 425 | * psci_idle_probe - Initializes PSCI cpuidle driver |
| 426 | * |
| 427 | * Initializes PSCI cpuidle driver for all present CPUs, if any CPU fails |
| 428 | * to register cpuidle driver then rollback to cancel all CPUs |
| 429 | * registration. |
| 430 | */ |
| 431 | static int psci_cpuidle_probe(struct faux_device *fdev) |
| 432 | { |
| 433 | int cpu, ret; |
| 434 | struct cpuidle_driver *drv; |
| 435 | struct cpuidle_device *dev; |
| 436 | |
| 437 | for_each_present_cpu(cpu) { |
| 438 | ret = psci_idle_init_cpu(dev: &fdev->dev, cpu); |
| 439 | if (ret) |
| 440 | goto out_fail; |
| 441 | } |
| 442 | |
| 443 | psci_idle_init_syscore(); |
| 444 | psci_idle_init_cpuhp(); |
| 445 | return 0; |
| 446 | |
| 447 | out_fail: |
| 448 | while (--cpu >= 0) { |
| 449 | dev = per_cpu(cpuidle_devices, cpu); |
| 450 | drv = cpuidle_get_cpu_driver(dev); |
| 451 | cpuidle_unregister(drv); |
| 452 | psci_cpu_deinit_idle(cpu); |
| 453 | } |
| 454 | |
| 455 | return ret; |
| 456 | } |
| 457 | |
| 458 | static struct faux_device_ops psci_cpuidle_ops = { |
| 459 | .probe = psci_cpuidle_probe, |
| 460 | }; |
| 461 | |
| 462 | static bool __init dt_idle_state_present(void) |
| 463 | { |
| 464 | struct device_node *cpu_node __free(device_node) = |
| 465 | of_cpu_device_node_get(cpu: cpumask_first(cpu_possible_mask)); |
| 466 | if (!cpu_node) |
| 467 | return false; |
| 468 | |
| 469 | struct device_node *state_node __free(device_node) = |
| 470 | of_get_cpu_state_node(cpu_node, index: 0); |
| 471 | if (!state_node) |
| 472 | return false; |
| 473 | |
| 474 | return !!of_match_node(matches: psci_idle_state_match, node: state_node); |
| 475 | } |
| 476 | |
| 477 | static int __init psci_idle_init(void) |
| 478 | { |
| 479 | struct faux_device *fdev; |
| 480 | |
| 481 | if (!dt_idle_state_present()) |
| 482 | return 0; |
| 483 | |
| 484 | fdev = faux_device_create(name: "psci-cpuidle" , NULL, faux_ops: &psci_cpuidle_ops); |
| 485 | if (!fdev) { |
| 486 | pr_err("Failed to create psci-cpuidle device\n" ); |
| 487 | return -ENODEV; |
| 488 | } |
| 489 | |
| 490 | return 0; |
| 491 | } |
| 492 | device_initcall(psci_idle_init); |
| 493 | |